Self-adaptive TBM hob rotating speed signal denoising method

Abstract

The invention discloses a self-adaptive TBM hob rotating speed signal denoising method, and relates to the technical field of TBM hobs, and the method comprises the following steps: employing ICEEMDAN to inhibit a mode aliasing phenomenon, and decomposing an original signal into intrinsic mode functions IMF of different scales; an IMF selection strategy based on kurtosis EK is developed to adaptively identify a noise dominant IMF, and random noise in the IMF is further eliminated through WSTD; reconstructing the de-noised IMF and the non-noise dominant IMF to obtain a de-noised signal; and data of four typical working conditions of normal, uniform abrasion, tipping and non-uniform abrasion are collected for analysis and verification. The adaptive TBM hob rotating speed signal denoising method shows a good denoising effect in a complex and strong noise environment, has robustness and adaptability to complex working conditions, and provides an effective data basis for wear state recognition and fault diagnosis of the TBM hob.

Description

Technical Field

[0001] This invention relates to the field of TBM hobbing technology, and in particular to an adaptive method for denoising the rotational speed signal of a TBM hobbing cutter. Background Technology

[0002] The cutter head is the core rock-breaking tool of a tunnel boring machine (TBM). Working long hours in hard rock environments, accurately sensing the cutter head's operating status is crucial for ensuring safe and efficient TBM tunneling. The cutter head rotates passively under the action of propulsion force and the cutterhead's motion, breaking rock through rolling. Its rotational speed directly reflects the cutter's condition. Figure 1 As shown, the fluctuations in the rotational speed signal characteristics are a true structural response to changes in the contact force between the cutter and the rock, containing information about the interaction process and are crucial for condition assessment. However, during tunneling, this signal is subject to strong noise and transient impact interference, and there is a significant conflict between denoising and preserving signal details, hindering the accurate extraction of high-quality signals.

[0003] Traditional fixed-parameter filters show significant inadequacy in adapting to highly nonlinear and rapidly changing signals; wavelet transform (WT) denoising is extremely sensitive to wavelet basis functions and threshold settings, and improper parameter selection can easily lead to excessive signal smoothing or residual noise; adaptive or improved threshold strategies (such as WSTD) still heavily rely on parameter adjustment and are difficult to fully adapt to the complex and ever-changing signal characteristics under rock breaking conditions.

[0004] In recent years, Empirical Mode Decomposition (EMD) and its variants have been able to decompose signals into a set of intrinsic mode functions (IMFs), but they suffer from severe mode aliasing problems. Noise-assisted algorithms such as Ensemble Empirical Mode Decomposition (EEMD) and Complete Ensemble Empirical Mode Decomposition (CEEMDAN) have improved decomposition accuracy, but still suffer from problems such as a large number of redundant modes, unclear signal-noise boundaries, and reliance on human experience. Variational Mode Decomposition (VMD) relies on a preset number of modes and penalty parameters, lacks an effective mode recognition mechanism, and may still erroneously delete useful information in complex non-stationary signals.

[0005] In summary, existing methods suffer from two major problems in the practical application of hob speed signal denoising: first, they lack an adaptive feature recognition mechanism, making it difficult to retain meaningful weak features in strong noise backgrounds; second, the reconstructed signal relies on manual parameter adjustment, failing to meet the high robustness requirements under complex working conditions. Therefore, there is an urgent need for a signal denoising method with strong impact adaptability, adaptive noise recognition, and fine feature preservation capabilities. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive method for denoising the rotational speed signal of a TBM hob, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides an adaptive TBM hobbing speed signal denoising method, comprising the following steps: S1. The original rotational speed signal is decomposed using the improved complete set empirical mode decomposition ICEEMDAN to obtain the intrinsic mode functions (IMFs) and residual terms at different scales, and the IMFs are sorted from high to low frequency. S2. Calculate the envelope kurtosis EK for each IMF, set an adaptive threshold based on the minimum EK value, and classify the IMFs into two categories: effective IMFs and noise-dominated IMFs. S3. Wavelet soft threshold denoising (WSTD) is used to denoise the noise-dominated IMF to obtain a denoised IMF, while the effective IMF remains unchanged. S4. Superimpose and reconstruct the denoised IMF and the effective IMF to obtain the denoised signal; S5. Build a scaled-down hob test platform and collect data from four typical working conditions: normal, uniform wear, chipping, and uneven wear for analysis and verification.

[0008] Preferably, the specific steps of S1 are as follows: S11. Iterate the original signal and the initial residual. In each iteration, generate 2 by adding symmetrical Gaussian white noise pairs. N Group noise disturbance signal: S12. Perform EMD decomposition on each group of noise disturbance signals, extract the first IMF, and calculate the average value of the first IMF for all samples: S13. Update the residuals and IMF average value. This process is iterated until the residuals show a monotonic trend or meet the preset termination criteria, and the original speed signal is decomposed.

[0009] Preferably, the specific formula for the first IMF average value in S12 is: ; in, IMF 1( t () represents the first IMF average. E 1() represents the first IMF obtained by EMD decomposition of the noisy signal. r 0 ±(i) ( t ) is the first i Group noise disturbance signal. N For the noise pair quantity.

[0010] Preferably, the specific formula for decomposing the original rotational speed signal in S13 is expressed as follows: ; in, x ( t () represents the original signal. r k ( t ) is the first k The residual is updated in each iteration. IMF k ( t ) is the first k The average IMF value is updated in each iteration.

[0011] Preferably, the specific steps of S2 are as follows: S21. Construct analytic signals using Hilbert transform; S22. Calculate the modulus of the analytic signal; S23. Calculate the envelope kurtosis EK; S23. Determine the noise-dominated IMF based on the adaptive threshold.

[0012] Preferably, the adaptive threshold in S23 is defined as: ; If the k The EK of the IMF, i.e., EK k >θ, then IMF k Marked as noise-dominant and denoised using wavelet thresholding; otherwise, IMF k It will be reserved for reconstruction; Baseline k 0 and dispersion s Defined by the 10th percentile and interquartile range (IQR): ; relative factor η It is data-adaptive and cuts within a certain range: ; Among them, clip( x , a , b )=min{max{ x , a}, b}

[0013] Preferably, the specific steps of S3 are as follows: S31. Perform wavelet decomposition on the noise-dominated IMF to obtain low-frequency approximation coefficients and high-frequency detail coefficients; S32. Apply a soft thresholding function to high-frequency detail coefficients; S33. Reconstruct the denoised IMF using inverse wavelet transform.

[0014] Preferably, the specific formula for the denoised IMF in S33 is as follows: ; in, For noise reduction of IMF, ɑ k These are low-frequency approximation coefficients. d k For high-frequency detail coefficients.

[0015] Preferably, the specific formula for superposition and reconstruction in S4 is as follows: ; in, For denoising signals, S 1 represents the effective IMF set. S 2 represents the noise-dominated IMF set.

[0016] Preferably, the analysis and verification in S5 includes the acquisition and analysis of the cutter speed signal, simulation experiments and analysis, and rock breaking experiments and analysis.

[0017] Therefore, the present invention employs the above-mentioned adaptive TBM hobbing speed signal denoising method, which has the following beneficial effects: (1) An adaptive denoising framework integrating ICEEMDAN, EK and WSTD was constructed, which takes into account the requirements of modal decoupling, impact recognition and detail preservation, and effectively addresses the problem of strong noise and weak features coexisting in the TBM hobbing speed signal.

[0018] (2) An EK-based IMF screening strategy is proposed to replace the traditional fixed threshold method, which realizes the adaptive separation of noise and feature components in non-stationary environments and improves the adaptability of the method under transient interference and complex working conditions.

[0019] (3) Experiments were conducted on the scaled-down hobbing platform under four typical cutter ring health conditions: normal, uniform wear, chipping, and uneven wear. This verified that the invention has robustness and adaptability in complex rock breaking conditions.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the cutter motion and rock-breaking mechanism in an embodiment of an adaptive TBM cutter speed signal denoising method of the present invention. (a) shows the cutter motion principle, (b) shows the cutter rock-breaking principle, and (c) shows the cutter speed signal characteristics under different working conditions. Figure 2This is a flowchart illustrating an embodiment of an adaptive TBM hobbing speed signal denoising method according to the present invention; Figure 3 This serves as an experimental platform for an embodiment of the adaptive TBM hobbing speed signal denoising method of the present invention. Figure 4 This is a module for acquiring the health status and rotational speed of a cutter ring, as an embodiment of an adaptive TBM hobbing speed signal denoising method of the present invention. Figure 5 This is a segment of rotational speed data measured by a gyroscope module in an embodiment of an adaptive TBM hobbing speed signal denoising method of the present invention. Figure 6 This is a synthesized signal curve diagram of an embodiment of the adaptive TBM hobbing speed signal denoising method of the present invention; Figure 7 The ICEEMDAN decomposition result of the synthesized signal in an embodiment of the adaptive TBM hob speed signal denoising method of the present invention; Figure 8 The comparison results of EK values ​​and adaptive thresholds for each IMF in an embodiment of the adaptive TBM hobbing speed signal denoising method of the present invention are shown. Figure 9 The following are the denoising results of different aspects of an adaptive TBM hobbing speed signal denoising method according to the present invention: (a) is the overall time domain denoising result, and (b) is the local denoising result within the blue marked area of ​​15-16 seconds. Figure 10 This is an offset stacking diagram of some original rotation speed signals collected during rock breaking of four types of cutter rings in 13s, according to an embodiment of the adaptive TBM cutter speed signal denoising method of the present invention. Figure 11 The ICEEMDAN decomposition result is shown in the embodiment of the adaptive TBM hob speed signal denoising method of the present invention. Figure 12 The EK values ​​of each intrinsic mode function (IMF) in an embodiment of the adaptive TBM hobbing speed signal denoising method of the present invention are shown. Figure 13 The denoising results of different methods under normal working conditions of the cutter ring are shown in the embodiment of the adaptive TBM hobbing speed signal denoising method of the present invention. Figure 14 This is an embodiment of an adaptive TBM hobbing speed signal denoising method according to the present invention. Figure 13 The magnified images of local signals are shown in Figure 1. (a) is a magnified image of the rock breaking stage, and (b) is a magnified image of the initiation stage. Figure 15 The signal processing results of five denoising methods under uniform wear conditions are presented in an embodiment of the adaptive TBM hobbing speed signal denoising method of the present invention. Figure 16 The following are the denoising results of different methods under the condition of cutter ring breakage in an embodiment of the adaptive TBM hobbing speed signal denoising method of the present invention: (a) is the overall time domain denoising result; (b) is the local denoising result in the blue marked area from 38.75 seconds to 38.95 seconds. Figure 17 The following are the denoising results of different methods under the condition of uneven wear of the cutter ring in an embodiment of the adaptive TBM hob speed signal denoising method of the present invention: (a) is the overall time domain denoising result; (b) is the local denoising result in the blue marked area within 38.1 seconds to 38.5 seconds. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0024] Example Please see Figures 1-17 This invention provides an adaptive denoising method for TBM hobbing speed signals. This method is based on an improved adaptive denoising method (ICEEMDAN-EK-WSTD) using Complete Ensemble Empirical Mode Decomposition (ICEEMDAN), Envelope Kujicism (EK), and Wavelet Soft Thresholding Denoising (WSTD), and includes the following steps: I. Noise Reduction Method for Hobbing Speed ​​Signal

[0025] 1. Use ICEEMDAN to decompose the original speed signal.

[0026] ICEEMDAN is an improved signal decomposition method for non-stationary complex signals. This method achieves effective signal decomposition by generating eigenmode functions (IMFs), resulting in lower mode aliasing and stronger interpretability. Furthermore, it eliminates the need for manually setting decomposition levels, avoiding the performance degradation associated with fixed-parameter methods (such as VMD).

[0027] Let the original signal be x(t), and the initial residual be denoted as r0(t). In each iteration, 2N sets of noise perturbation signals are generated by adding symmetrical white Gaussian noise pairs: ; in, ω (i) ( t ) represents the Gaussian white noise added in the i-th iteration. σ 0 represents the noise amplitude. N This represents the number of noise pairs (integration count).

[0028] For each group of noise signals r 0 ±(i) EMD decomposition was performed to extract the first IMF. The final first IMF was calculated by averaging all samples. ; in, E 1() represents the first IMF obtained by EMD decomposition of the noisy signal.

[0029] Then update the residual: ; This process iterates continuously. For the k-th iteration, the residual is updated as follows: ; The final residual is updated as follows: ; The iterative process continues until the residual is found. r K ( t The signal continues until it exhibits a monotonic trend or meets a preset termination criterion. The final signal decomposition is expressed as: ; Compared with traditional EMD and CEEMDAN methods, ICEEMDAN makes two key improvements: (1) it adds symmetrical paired Gaussian white noise in each iteration to suppress mode aliasing and reduce residual noise; (2) it achieves complete signal reconstruction without manual parameter tuning. Therefore, ICEEMDAN is particularly suitable for processing hob rotation speed signals containing random impacts and high-frequency vibration noise. In this embodiment, ICEEMDAN is used to decompose the original rotation speed signal acquired by the sensor, and the decomposed IMF provides the basis for subsequent noise identification and denoising processing.

[0030] 2. Calculate the envelope kurtosis EK and classify the IMFs into effective IMFs (set S1) and noise-dominated IMFs (set S2).

[0031] After obtaining the IMF, it is necessary to determine which IMFs contain noise components. Noise-dominated IMFs usually have irregular envelopes and high kurtosis values. Therefore, EK can be regarded as an important criterion for determining whether an IMF is noise-dominated.

[0032] For each IMF component obtained from the decomposition, calculate its EK value, and set an adaptive threshold based on the minimum EK value. θ The IMFs are compared with an adaptive threshold. IMFs exceeding the threshold are marked as noise-dominated and require further denoising processing. The specific steps are as follows: Let IMF be IMF k (t), constructing an analytic signal using the Hilbert transform: ; in, For Hilbert transform operators, j It is the imaginary unit.

[0033] Traditional energy or entropy-based evaluation metrics have certain limitations in distinguishing between effective transient signals and random impulse noise, and often fail to accurately identify transient impulse characteristics. In this embodiment, EK is a higher-order statistic based on the signal envelope after Hilbert transform, which can effectively identify impulse characteristics in the signal.

[0034] The envelope of an IMF signal is the modulus of the analytic signal. ; Envelope kurtosis calculation can be expressed as: ; in, The mean of the envelope. It expresses expectation.

[0035] The adaptive threshold is defined as: ; If the k The EK of the IMF, i.e., EK k >θ, then IMF k It is marked as noise-dominant and denoised by wavelet thresholding; otherwise, it will be retained for reconstruction. To avoid degenerate decisions, we always retain at least one IMF (the one with the smallest EK).

[0036] Baseline k 0 and dispersion s Defined by the 10th percentile and interquartile range (IQR): ; relative factor η It is data-adaptive and cuts within a certain range: ; Among them, clip( x , a , b )=min{max{ x , a}, b}

[0037] 3. Wavelet soft thresholding (WSTD) is used to denoise the noise-dominant IMF.

[0038] Wavelet decomposition of the IMF yields low-frequency approximation coefficients. ɑ k and high frequency detail coefficient d k : ; High-frequency portion only d k Applying the soft threshold function: ; The threshold λ is adaptively calculated based on the median absolute deviation (MAD) of the detail coefficients.

[0039] Reconstruct the denoised signal using inverse wavelet transform: ; This method effectively suppresses high-frequency noise while preserving the transient characteristics of the signal, achieving a good balance and thus improving signal quality.

[0040] 4. Overlay all valid IMFs with the denoised IMFs to reconstruct the final denoised signal.

[0041] ; The "decomposition-classification-denoising-reconstruction" strategy suppresses noise while preserving key transient features, making it particularly suitable for non-stationary signal processing in TBM rock-breaking environments.

[0042] II. Experiment and Analysis.

[0043] 1. Acquisition and analysis of hobbing speed signals.

[0044] like Figure 3 As shown, the experimental platform mainly consists of a hydraulic system, a feed platform, a control system, and a tapered cutting cutter, supporting control of the cutter's penetration depth, lateral displacement, and rock-breaking speed. The rock sample used was sandstone with a uniaxial compressive strength of 50 MPa.

[0045] The scaled-down roller cutter developed in this embodiment is designed with a load of 50kN and can be equipped with a prefabricated scaled-down cutter ring to simulate the roller cutting process under different wear conditions on a linear rock breaking test bench. Figure 4 The image shows the rotational speed acquisition system and the cutter ring in four health states: normal, uniform wear, chipped edge, and uneven wear. Two rotational speed data acquisition systems are used. (1) Encoder speed measurement system: The encoder rotates through a 5:1 gear set linked with the hob. The speed signal measured by the encoder is transmitted to the computer via an RS485–USB module. The sampling rate is 50Hz and the resolution is 0.003°.

[0046] (2) Gyroscope module: Installed inside the hob, the rotation speed data is transmitted wirelessly and then transmitted to the computer via a wireless receiving module. The relevant parameters of the gyroscope module are: sampling rate 400Hz, range 2000dps, maximum transmission speed 40KB / s, and average operating current <1mA.

[0047] like Figure 5 The image shows a segment of rotational speed data measured by the gyroscope module. The three background colors represent the three stages of the signal: stable rock-breaking, strong impact rock-breaking, and unstable disturbance. The stable rock-breaking stage is shown in detail in the enlarged image. In this stage, the broken rock fragments are small and uniform, and the cutter is in close contact with the rock. Low-frequency fluctuations in rotational speed are clearly visible in the image, along with significant spike-like noise. In the strong impact rock-breaking stage, the rotational speed changes drastically. The broken rock fragments are larger, and the contact force between the cutter and the rock changes significantly, resulting in greater rotational speed fluctuations. In the unstable disturbance stage, the cutter loses contact with the rock, causing the rotational speed to rapidly drop to zero. It can be seen that the cutter-rock interaction makes the rotational speed variation complex and introduces significant signal noise. Accurately filtering out rotational speed noise while retaining more detailed rotational speed information is crucial for subsequent wear prediction and fault diagnosis of the cutter based on rotational speed.

[0048] 2. Simulation experiment and analysis.

[0049] Since noise is inevitably introduced into the measured signal, the true rotational speed value is difficult to obtain directly. Although the data measured by the encoder can serve as a reference benchmark for the gyroscope module data, in order to more accurately evaluate the performance of the proposed denoising method, this embodiment constructs a synthetic rotational speed signal based on the characteristics of the experimentally measured rotational speed data, and quantitatively analyzes the denoising performance of the algorithm.

[0050] The signal is divided into two parts: a pure rotational speed signal and various types of noise. The constructed signal comprehensively considers the effects of machine vibration, impact interference, and internal sensor noise, and is modeled as the sum of four parts: ; in, ω pure (t) represents the pure hobbing speed signal. η vib (t) represents high-frequency mechanical vibration noise. η imp (t) represents the impact disturbance caused by the contact between the cutter and the rock. η white (t) is zero-mean Gaussian white noise.

[0051] Pure speed signal ω pure (t) includes a constant baseline velocity ω 0 and the low-frequency fluctuation term caused by the knife-rock contact: ; in, T Indicates the total duration of the signal. T 0 represents the unloaded period before and after rock breaking; a sine wave simulates the low-frequency contact fluctuations; A fluc The amplitude of low-frequency fluctuations. f fluc It represents the frequency of low-frequency fluctuations.

[0052] Vibration noise η vib (t) Simulates high-frequency disturbances within a mechanical system: ; in, A vib and f vib These represent the vibration amplitude and frequency, respectively.

[0053] Impact interference signal η impModeling (t) is particularly crucial. While conventional impact models (such as exponentially decaying cosine) can reflect the trend of free vibration, they are difficult to accurately simulate the local, high-frequency, short-time disturbances generated by the TBM cutter impact. Therefore, this embodiment draws on the idea of ​​Gabor-type functions to model the impact as a sinusoidal pulse modulated by a Gaussian envelope, in order to capture the local high-energy disturbance characteristics during the cutter impact process: ; in, t k For the first k The central moment of the secondary impact (within the interval [ T 0, T - T (randomly selected from within 0) A imp For the impact amplitude, σ imp Control the pulse width, f ring This is the internal oscillation frequency.

[0054] Sensor internal noise η white (t) Simulated as zero-mean Gaussian white noise: ; in, σ white To control the pulse width.

[0055] The aforementioned analog signals and their denoising were all performed on a unified computing platform, such as... Figure 6 As shown, the synthesized signal curves are displayed, including the clean signal and the noisy signal.

[0056] After signal synthesis, the synthesized noisy signal is adaptively decomposed using the ICEEMDAN method, with 50 integration iterations and a noise amplitude of 0.02. Figure 7 The image shows the ICEEMDAN decomposition results of the synthesized signal. The EK value is calculated for each IMF generated by the signal decomposition and compared with a threshold. Figure 8 As shown, the comparison results of the EK value of each IMF and the adaptive threshold (set to 1.5 times the minimum EK) are presented. It can be seen that IMF1 to IMF8 exceed the threshold and are therefore determined to be noise-dominated IMFs, while the other IMFs are determined to be valid IMFs. WSTD denoising is performed on the noise-dominated IMFs, and then the denoised IMFs and valid IMFs are reconstructed to restore the hob rotation speed signal.

[0057] To verify the effectiveness of the method, this embodiment compares the proposed ICEEMDAN-EK-WSTD method with four denoising methods: Method 1: TCO-WD-BWF: Wavelet denoising and Butterworth low-pass filtering are used, and the parameters are optimized by the Tornado Algorithm (TCO). Method 2: MVMD-SE&PE-WSTD: The signal is decomposed using multivariate variational mode decomposition (MVMD). The IMF is classified by sample entropy (SE) and permutation entropy (PE). Low-entropy components are retained, and the rest are denoised using WSTD. Method 3: EMD-EK-WSTD: The signal is decomposed using EMD, and IMFs are classified based on EK. Noise-dominated IMFs are denoised using WSTD. Method 4: ICEEMDAN-EK: Apply ICEEMDAN and EK classification, and discard noisy IMFs directly without processing.

[0058] To quantitatively evaluate the denoising performance of denoising methods, two standard metrics were used: signal-to-noise ratio (SNR) and root mean square error (RMSE). SNR assesses the degree of noise suppression, while RMSE measures the deviation from the true signal.

[0059] ; ; in, x ( i () is the ideal signal. The signal after denoising. N This represents the number of sampling points.

[0060] like Figure 9 As shown, the denoising results of the above methods are presented, and Table 1 below summarizes the evaluation metrics. The ICEEMDAN-EK method performs the worst (SNR: 20.94dB, RMSE: 0.366) due to severe loss of detail caused by directly discarding the IMF. TCO-WD-BWF has good overall performance in terms of smoothness and accuracy, and has the shortest computation time (0.58s), but still suffers from detail distortion. MVMD-SE & PE-WSTD have slightly higher residual noise. EMD-EK-WSTD retains more details, but performs poorly under transient impact. The proposed ICEEMDAN-EK-WSTD method performs best overall, achieving the highest SNR and lowest RMSE. Although the computation time is slightly longer (24.3s), it is completely acceptable in offline processing scenarios.

[0061] Table 1 Evaluation of various methods 3. Rock breaking experiment and analysis.

[0062] Simulation experiments have demonstrated the excellent noise reduction performance of this invention. To further verify the robustness and adaptability of this method under actual working conditions, this embodiment constructed a linear rock-breaking experimental platform and conducted rolling rock-breaking experiments with four different cutter rings: normal, uniform wear, chipped edge, and uneven wear. The normal cutter ring had a diameter of 220 mm, while the uniform wear cutter ring had a radius 2 mm smaller than the normal cutter ring. The chipped edge cutter ring had 18 evenly distributed 6 mm × 8 mm notches, and the uneven wear cutter ring exhibited a pair of 1 mm, 2 mm, and 3 mm uneven wear characteristics on its cutting edge.

[0063] like Figure 10 The image shows a stacked offset of some original rotational speed signals collected during 13 seconds of rock breaking with four different cutter rings. From top to bottom, the signals represent the rotational speed signals of uneven wear, chipped edges, uniform wear, and normal cutter rings. The four types of signals exhibit distinct characteristics. Unevenly worn cutter rings show significant signal fluctuations when passing through the fault point, with a long duration and large amplitude. Chipped edges show rapid changes in rotational speed with small amplitude changes when passing through the fault point. Uniformly worn and normal cutter rings have similar signal characteristics; the main difference lies in the average rotational speed. The sudden drop in these two rotational speed signals is due to pitting in the rock, causing the cutter to temporarily lose contact with the rock and experience a brief stall. These characteristics of different working conditions provide important basis for subsequent prediction of cutter ring wear status and fault diagnosis based on rotational speed signals.

[0064] In the signal processing, taking the rotational speed signal of a normal hobbing cutter as an example, the ICEEMDAN algorithm is first used to decompose the signal (ensemble count 50, noise amplitude 0.02), resulting in 13 IMFs. The decomposition results are as follows: Figure 11 As shown in the figure. Then, the envelope kurtosis (EK) of each IMF is calculated and compared with an adaptive EK threshold. IMFs exceeding the threshold are identified as noise-dominant components. The results are shown in the figure. Figure 12 As shown, the EK values ​​of IMF1-IMF9 exceed the threshold and are marked as noise-dominated IMFs. The WSTD method is used to denoise the noise-dominated IMFs, and the denoised signal is finally reconstructed.

[0065] The five denoising methods compared in the simulation experiment were also applied to this set of experimental data, and the results are as follows: Figure 13 As shown, the overall trend of denoising results from all methods is basically the same, but there are differences in detail preservation. The ICEEMDAN-EK method outputs an overly smooth signal with significant lag. To more clearly compare the detail performance of each method, as shown... Figure 14As shown, local signals in the early and middle stages of rock breaking were magnified. From the two magnified images, it can be seen that TCO-WD-BWF and MVMD-SE&PE-WSTD achieve a high degree of smoothing during processing, resulting in the elimination of some key peak and valley features; especially in the early stages of rock breaking, TCO-WD-BWF even exhibits unreasonable data fluctuations. EMD-EK-WSTD and the proposed method, on the other hand, can better preserve these detailed changes. However, EMD-EK-WSTD still has sharp peak artifacts after denoising, and its overall smoothness is slightly worse.

[0066] Table 2 below presents the quantitative evaluation results of different methods under normal tool ring experiments. The ICEEMDAN-EK-WSTD method achieved the highest signal-to-noise ratio (SNR: 29.13 dB) and the lowest root mean square error (RMSE: 0.1017), demonstrating the robustness of the proposed method. The proposed algorithm has a computation time of 17.2 seconds, significantly better than EMD-EK-WSTD's 25.7 seconds. Although it takes much longer than TCO-WD-BWF and MVMD-SE&PE-WSTD, this time cost is perfectly acceptable given the experimental acquisition cycle of 30 seconds.

[0067] Table 2. Quantitative evaluation results of different methods under normal blade ring experiments. Table 2 also shows the denoising effects under the other three cutter ring experiments. The results under the four conditions are highly consistent, fully demonstrating the adaptability of the proposed ICEEMDAN-EK-WSTD method to different rock breaking conditions. This method achieves an average SNR of 29.12 dB and an RMSE of 0.1381 under all conditions, representing an average improvement of 3.35 dB (maximum improvement of 10.73 dB) compared to other comparative methods. Figures 15-17 As shown, the noise reduction effects of three types of cutter rings—uniform wear, chipping, and non-uniform wear—are demonstrated in turn, further illustrating the comprehensive advantages of the proposed method in terms of signal fidelity and noise suppression.

[0068] like Figure 15 The image shows the signal processing results of five denoising methods under the condition of "uniformly worn cutter ring". The overall speed signal under this condition is relatively stable, but it still contains significant high-frequency noise and several instantaneous speed troughs. These troughs are caused by pits on the rock surface, resulting in a brief loss of contact by the hob during cutting, leading to a sudden drop in speed. These fluctuations have practical physical significance and should therefore be preserved as much as possible during signal processing. However, the ICEEMDAN-EK method incorrectly deleted this segment during denoising.

[0069] As shown in the enlarged area, the ICEEMDAN-EK-WSTD method proposed in this embodiment effectively suppresses high-frequency noise while maintaining the overall trend and preserving physically meaningful valleys. The rotational speed signal becomes smoother without losing the basic fluctuations caused by the knife-rock interaction. The EMD-EK-WSTD method, due to the addition of wavelet denoising, performs better than ICEEMDAN-EK, but it is based on traditional EMD decomposition and exhibits some mode aliasing, resulting in slight distortion in short-term transients and significant signal baseline fluctuations. The MVMD-SE&PE-WSTD method has a moderate denoising effect, but waveform distortion still exists. Although the TCO-WD-BWF method can generate a smooth signal, over-smoothing not only eliminates noise but also smooths out some engineering-significant fluctuations, especially the signal valleys.

[0070] Overall, under these slight impact characteristics, the present invention achieves the best balance between noise suppression and preservation of effective transient information, and is suitable for signal processing under slight disturbance conditions such as "uniform wear of the cutter ring".

[0071] like Figure 16 and Figure 17 As shown, the noise reduction performance under two complex working conditions—"edge chipping" and "uneven wear"—is further analyzed. These types of cutter rings experience significant speed abrupt changes and aperiodic disturbances during cutting due to structural defects (such as edge breakage or uneven wear). For example... Figure 16 As shown, the "broken blade ring" causes a sudden drop in rotational speed due to edge notches, resulting in a clear fault point. Although the signal is accompanied by strong transients and high-frequency noise, the method in this embodiment can effectively suppress background noise while preserving the signal drop.

[0072] As can be seen from the magnified image, the present invention accurately reproduces the amplitude and shape of the instantaneous speed change caused by the fault, while other methods have different degrees of error: the ICEEMDAN-EK, TCO-WD-BWF and MVMD-SE&PE-WSTD methods over-smooth the fault point, resulting in weakened fluctuations; the EMD-EK-WSTD method retains the abrupt change point, but has more residual noise.

[0073] like Figure 17 The image shows the processing results under "unevenly worn cutter ring," where the cutter's uneven wear and contact with the rock cause multiple aperiodic disturbances. The signal contains several abrupt changes, with typical "failure points" marked in the figure, caused by severe cutter ring eccentricity. The proposed method maintains high adaptability under this complex background: it not only successfully suppresses high-frequency interference but also accurately preserves the signal morphology and energy near the failure point. As seen in the magnified image, the rising and falling edges of the signal near the failure point are completely preserved, without significant waveform distortion or over-filtering.

[0074] This embodiment retains the low-frequency fluctuations, transient velocity valleys, and fault mutation characteristics reflecting the cutter-rock interaction in the signal, while effectively suppressing high-frequency spikes and random pulse noise introduced by rock spalling impact, vibration, and sensor electromagnetic interference. In the experiment, the reconstructed signal successfully preserved key information such as sudden speed drops under "edge chipping" conditions and disturbance waveforms under "uneven wear" conditions, improving the signal's readability and physical interpretability, and providing reliable raw data support for hob wear trend modeling and fault identification algorithms.

[0075] Therefore, this invention employs an adaptive TBM hob speed signal denoising method, which demonstrates excellent noise reduction performance in complex, high-noise environments, providing an effective data foundation for TBM hob wear condition identification and fault diagnosis. Future research will further explore the application potential of this method in multi-sensor signal fusion and complex working condition identification.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An adaptive method for denoising the rotational speed signal of a TBM hob, characterized in that, Includes the following steps: S1. The original rotational speed signal is decomposed using the improved complete set empirical mode decomposition ICEEMDAN to obtain the intrinsic mode functions (IMFs) and residual terms at different scales, and the IMFs are sorted from high to low frequency. S2. Calculate the envelope kurtosis EK for each IMF, set an adaptive threshold based on the minimum EK value, and classify the IMFs into two categories: effective IMFs and noise-dominated IMFs. S3. Wavelet soft threshold denoising (WSTD) is used to denoise the noise-dominated IMF to obtain a denoised IMF, while the effective IMF remains unchanged. S4. Superimpose and reconstruct the denoised IMF and the effective IMF to obtain the denoised signal; S5. Build a scaled-down hob test platform and collect data from four typical working conditions: normal, uniform wear, chipping, and uneven wear for analysis and verification.

2. The adaptive TBM hobbing speed signal denoising method according to claim 1, characterized in that, The specific steps of S1 are as follows: S11. Iterate the original signal and the initial residual. In each iteration, generate 2 by adding symmetrical Gaussian white noise pairs. N Group noise disturbance signal: S12. Perform EMD decomposition on each group of noise disturbance signals, extract the first IMF, and calculate the average value of the first IMF for all samples: S13. Update the residuals and IMF average value. This process is iterated until the residuals show a monotonic trend or meet the preset termination criteria, and the original speed signal is decomposed.

3. The adaptive TBM hobbing speed signal denoising method according to claim 2, characterized in that, The specific formula for the first IMF average value in S12 is as follows: ； in, IMF 1( t () represents the first IMF average. E 1() represents the first IMF obtained by EMD decomposition of the noisy signal. r 0 ±(i) ( t ) is the first i Group noise disturbance signal. N For the noise pair quantity.

4. The adaptive TBM hobbing speed signal denoising method according to claim 2, characterized in that, The specific formula for decomposing the original rotation speed signal in S13 is expressed as follows: ； in, x ( t () represents the original signal. r k ( t ) is the first k The residual is updated in each iteration. IMF k ( t ) is the first k The average IMF value is updated in each iteration.

5. The adaptive TBM hobbing speed signal denoising method according to claim 1, characterized in that, The specific steps of S2 are as follows: S21. Construct analytic signals using Hilbert transform; S22. Calculate the modulus of the analytic signal; S23. Calculate the envelope kurtosis EK; S23. Determine the noise-dominated IMF based on the adaptive threshold.

6. The adaptive TBM hobbing speed signal denoising method according to claim 5, characterized in that, The adaptive threshold in S23 is defined as follows: ； If the k The EK of the IMF, i.e., EK k >θ, then IMF k Marked as noise-dominant and denoised using wavelet thresholding; otherwise, IMF k It will be reserved for reconstruction; Baseline k 0 and dispersion s Defined by the 10th percentile and interquartile range (IQR): ； relative factor η It is data-adaptive and cuts within a certain range: ； Among them, clip( x , a , b )=min{max{ x , a }, b } 7. The adaptive TBM hobbing speed signal denoising method according to claim 1, characterized in that, The specific steps of S3 are as follows: S31. Perform wavelet decomposition on the noise-dominated IMF to obtain low-frequency approximation coefficients and high-frequency detail coefficients; S32. Apply a soft thresholding function to high-frequency detail coefficients; S33. Reconstruct the denoised IMF using inverse wavelet transform.

8. The adaptive TBM hobbing speed signal denoising method according to claim 7, characterized in that, The specific formula for the denoised IMF in S33 is as follows: ； in, For noise reduction of IMF, a k These are low-frequency approximation coefficients. d k For high-frequency detail coefficients.

9. The adaptive TBM hobbing speed signal denoising method according to claim 1, characterized in that, The specific formula for superposition and reconstruction in S4 is as follows: ； in, For denoising signals, S 1 represents the effective IMF set. S 2 represents the noise-dominated IMF set.

10. The adaptive TBM hobbing speed signal denoising method according to claim 1, characterized in that: The analysis and verification in S5 includes the acquisition and analysis of the cutter speed signal, simulation experiments and analysis, and rock breaking experiments and analysis.

Images